Stable Layers Released: Decomposing Images into Editable RGBA Layers
bdsqlsz · x · 2026-08-11
Stability AI Labs has released the weights for the Stable Layers model, which decomposes a single RGB image into a stack of back-to-front RGBA layers (background and object layers), drastically improving image editability and compositing flexibility.
The model is an RL finetune of Qwen-Image-Layered, trained using Qwen 3.5 9b as the VLM teacher.
Requirements & Recommended Settings:
- Requires an 80GB-class GPU (e.g., A100-80 / H100 / H200) for comfortable inference.
- Recommended settings include the Heun sampler, 50 steps, CFG 1.0, at 640 px resolution to generate 4 layers.
More from Multimodal
- One Stylus Tap Turns Raw Ingredients Into Gourmet Ramen Using AI Video — SimplyAnnisa · 2026-08-11
- How to Identify AI Video Models by Their Default Duration — cocktailpeanut · 2026-08-11
- AI video frontier shifts from 5-second clips to native 30-second takes — Independent-Date393 · 2026-08-11
- Wan 3.0 beta can turn PDFs and slide decks directly into videos — Few-Profession421 · 2026-08-11
- Practical Guide to Training LoRA for Ace Step XL Audio Model — CryptoChangeling69 · 2026-08-11
- Alibaba Open-Sources Wan-Animate-2: Multi-Character & Real-Time Streaming — camenduru · 2026-08-11